A Novel Algorithm for Estimating the Sand Dune Density of the Taklimakan Desert Based on Remote Sensing Data

The dune density is an important parameter for representing the characteristics of desert geomorphology, providing a precise depiction of the undulating topography of the desert. Owing to the limitations of estimation methods and data availability, accurately quantifying dune density has posed a significant challenge; in response to this issue, we propose an innovative model to estimate dune density using a dune vertex search combined with four-directional orographic spectral decomposition. This study reveals several key insights: (1) Taklimakan Desert distributes approximately 5.31 × 107 dunes, with a linear regression fit R2 of 0.79 between the estimated and observed values. The average absolute error and root mean square error are calculated as 25.61 n/km2 and 30.48 n/km2, respectively. (2) The distribution of dune density across the eastern, northeastern, southern, and western parts of the Taklimakan Desert is relatively lower, while there is higher dune density in the central and northern areas. (3) The observation data constructed using the improved YOLOv8s algorithm and remote sensing imagery effectively validate the estimation results of dune density. The new algorithm demonstrates a high level of accuracy in estimating sand dune density, thereby providing crucial parameters for sub-grid orographic parameterization in desert regions. Additionally, its application potential in dust modeling appears promising.